SSH authentication shim for git-lfs.
There are several reasons you might want to use this despite the fact that the spec recommends against using SSH:
- You want to auth against a different host than your LFS server
- You want to have password-less authentication
- Re-using existing SSH keys infrastructure
- Not wanting to store unencrypted passwords in memory (credentials cache)
- You want to have authorisation based on LDAP group membership
- Re-using existing RBAC infrastructure
Build (go build main.go ldap.go; you may need to go get gopkg.in/ldap.v2 and go get gopkg.in/ini.v1 to resolve deps fist) and put somewhere inside PATH (e.g. /usr/local/bin/).
By default git-lfs-authenticate reads its configuration data from /etc/git-lfs-authenticate.conf. You can override this by providing the GIT_LFS_AUTHENTICATE_CONFIG environment variable.
This repository contains a sample config file in example.conf. Check it out.
Below is a short explanation of every option:
-
Lfs.Url — address of LFS server in URL format
-
Lfs.User — shared username that will be used to talk to LFS on behalf of the user
-
Lfs.Password — password for the shared user
-
Ldap.Urls — comma–separated list of LDAP servers in URL format
-
Ldap.Groups — comma–separated list of LDAP groups
-
Ldap.Cacert — path to cacert.pem (for when using TLS)
-
Ldap.Base — base DN for LDAP searches
LDAP servers are tried one–by–one in randomised order until either a match is found or a fatal error occurs (e.g. user not found).
The list of LDAP groups is OR-ed, i.e. a membership in at least one of the groups is sufficient.
The cacert file is optional. It’s still possible to use TLS without it but without hostname verification.
It is recommended that the system user used to talk to LFS is made low privileged, not being able to log into your other servers and run commands there.
This command will be invoked via the git-lfs client. See the spec for more details.